MiniMax Just Dropped a Open Source Seedance Killer @TheoreticallyMedia
MiniMax Just Dropped a Open Source Seedance Killer  @TheoreticallyMedia
Uploaded July 2026 | Updated November 2026, 13 hours ago
Today's video is sponsored by ElevenLabs! Try ElevenAgents and start chatting with your characters — plus grab 10,000 free credits courtesy of ElevenLabs right here: try.elevenlabs.io/w5183uo8pydc

This is a surprise: MiniMax is BACK. After nine months of silence since Hailuo 2.3, they’ve launched Hailuo 03 (aka MiniMax H3) — a fully multimodal AI video model with native audio, up to 15-second generations at 2K, and an Omni mode that takes up to 12 image references plus video and audio inputs. Is it the Seedance killer people are claiming? I ran it through the full gauntlet to find out.

In this video, we test Hailuo H3 across cinematic text-to-video (including dialogue, acting range, and multilingual support), image-to-video against the old 2.3 model, first frame/last frame, video extensions, and multi-reference character team-ups — plus what I’m seeing on pricing, which looks like it comes in around 50% under Seedance 2.0.

Key takeaways: dialogue and lip sync are shockingly good, fight choreography trades speed for stability (no morphing, no teleporting), prompts cap out at a generous 7,000 characters, and the Omni model takes your storyboards very seriously — maybe more seriously than Seedance does.

Note: H3 is now Open Source (couldn't say that when the video first came out) and is available over on Hugging Face:
huggingface.co/MiniMaxAI/MiniMax-H3

CHAPTERS

00:00 Intro: MiniMax Is Back
00:49 What's New in Hailuo 03
01:44 Dialogue & Acting Test (Twin Peaks Diner)
02:21 Cinematic Text-to-Video Tests
03:02 The Mob Movie Test
03:47 Multilingual Dialogue (Samurai Test)
04:39 Fight Scenes + the 7,000-Character Prompt Limit
06:10 Image-to-Video: Nine Months of Progress
07:03 Flamethrower Girl Returns
07:50 First Frame / Last Frame
08:30 The Omni Model: Multi-Reference
09:37 ElevenLabs Sponsor Segment
13:34 Sniper Girl One-Shot
14:13 Video References & Extensions
15:18 Sniper + Flamethrower Team-Up
16:02 "Reference Maxing" & Storyboards
17:25 Pricing Ballpark
18:23 Final Thoughts
MiniMax Just Dropped a Open Source Seedance KillerGenie 4 Is the FUTURE of AI World Models!OpenAI Cooked with Dots, But What Will It Cost You?Seedance 2.5 Just Got a Load of Awesome New Features!OpenAIs Stunning Image Model Has Some Cool Tricks!Ultimate NanoBanana Deep Dive (Hacks, Tips, and Tricks You Need To Know!)Googles FREE AI Image Game Changer Is KILLERGoogle Quietly Launched Its Best AI Video Tools (& Didnt Tell You)Veo-2 Drops on LTX Studio (Massive Price break!)AI Video Just Stepped Into the Real World!The 10 BEST AI Image Generators!Luma Is The BEST AI Image To Video Now? Plus: Pika & Google Updates!
Theoretically Media |

MiniMax Just Dropped a Open Source "Seedance Killer"

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER